Use of an artificial neural network algorithm and cokriging method for reservoir porosity modeling

Abstract

© SGEM2019. As a rule, well-log porosity is used as a reference information in the process of reservoir porosity model construction. At the same time, the regions of the model, where the porosity values are obtained by interpolating the well data, are characterized by a decrease in detail with distance from the wells, as well as by uncertainty in the interwell space. The uncertainty arising from interpolation can be significantly reduced as a result of the integrated use of well logging and 3D seismic data, and the usefulness of seismic data can be significantly enhanced through the use of neural network modeling, attribute analysis and geostatistics. Having an initial dataset of well-log porosity, based on the statistical evaluation it is possible to determine the set of seismic attributes that are most correlated with porosity by well logging. Then, the selected set of attributes is used to obtain the relation for converting them into porosity predicted by neural network algorithms. Thus, the above-mentioned steps allow to perform seismic porosity prediction by neural network modeling. Use of cokriging method allows using the predicted seismic porosity as a trend to adjust the porosity model constructed by well logging data, thereby clarifying its interwell distribution. At the same time, the cokriging method includes the use of two data sets (well logging data and predicted seismic porosity), and depending on the selection of cokriging coefficient it is possible to adjust the contribution made by each data set to the resulting porosity model.